融合异常检测与随机森林的微博转发行为预测方法

发布时间：2019-05-17 22:49

【摘要】：针对目前微博转发行为预测具有的特征选择任意性、准确率不高的问题,提出了融合异常检测与随机森林的微博转发行为预测方法。首先,提取用户基本特征、博文基本特征、博文内容主题特征,并基于相对熵计算用户活跃度、博文影响力;其次,通过结合过滤式与封装式特征选择方法筛选出关键特征组;最后,融合异常检测与随机森林算法,依据筛选后的关键特征组进行微博转发行为预测,并利用袋外数据误差估计设置随机森林中的决策树和特征数。在真实新浪微博数据集上与基于逻辑回归、决策树、朴素贝叶斯、随机森林等算法的微博转发行为预测方法进行实验对比,结果表明所提方法的预测准确率(90.5%)高于基准方法中最优的随机森林方法的预测准确率,同时验证了特征筛选方法的有效性。
[Abstract]:In order to solve the problem of arbitrary feature selection and low accuracy in Weibo forwarding behavior prediction, a Weibo forwarding behavior prediction method combining anomaly detection and random forest is proposed. Firstly, the basic features of users and the subject features of blog content are extracted, and the user activity and blog influence are calculated based on relative entropy. Secondly, the key feature groups are selected by combining filtering and encapsulation feature selection methods. Finally, combining anomaly detection and random forest algorithm, Weibo forwarding behavior is predicted according to the selected key feature groups, and the decision tree and feature number in random forest are set up by using out-of-bag data error estimation. The Weibo forwarding behavior prediction method based on logical regression, decision tree, naive Bays, random forest and other algorithms is compared with the Weibo forwarding behavior prediction method on the real Sina Weibo dataset. The results show that the prediction accuracy of the proposed method (90.5%) is higher than that of the optimal stochastic forest method in the benchmark method, and the effectiveness of the feature screening method is verified.
【作者单位】：桂林电子科技大学计算机与信息安全学院;桂林电子科技大学广西可信软件重点实验室;
【基金】：广西科技攻关项目(桂科攻1598019-6)资助
【分类号】：TP391.1;TP393.092

【参考文献】